Gas Sensing Properties of Perovskite Decorated Graphene at Room Temperature

This paper explores the gas sensing properties of graphene nanolayers decorated with lead halide perovskite (CH3NH3PbBr3) nanocrystals to detect toxic gases such as ammonia (NH3) and nitrogen dioxide (NO2). A chemical-sensitive semiconductor film based on graphene has been achieved, being decorated with CH3NH3PbBr3 perovskite (MAPbBr3) nanocrystals (NCs) synthesized, and characterized by several techniques, such as field emission scanning electron microscopy, transmission electron microscopy and X-ray photoelectron spectroscopy. Reversible responses were obtained towards NO2 and NH3 at room temperature, demonstrating an enhanced sensitivity when the graphene is decorated by MAPbBr3 NCs. Furthermore, the effect of ambient moisture was extensively studied, showing that the use of perovskite NCs in gas sensors can become a promising alternative to other gas sensitive materials, due to the protective character of graphene, resulting from its high hydrophobicity. Besides, a gas sensing mechanism is proposed to understand the effects of MAPbBr3 sensing properties.


Introduction
Atmospheric contamination is one of the most important environmental issues in current societies. According to the World Health Organization (WHO), air pollution is linked to 7 million premature deaths in 2012 [1]. The most common anthropogenic pollution sources are coming from industrial processes, automotive and energy production [2]. The gases emitted can produce an important greenhouse effect such as carbon dioxide (CO 2 ), methane (CH 4 ), and ozone (O 3 ), or can be dangerous for human health above very low levels of exposure. Nitrogen dioxide (NO 2 ) and ammonia (NH 3 ) are two examples of these harmful species. For these reasons, the in-field monitoring of harmful gases is a mandatory goal to control dangerous levels of contaminants. Above the threshold limits, when human health is at risk, gas sensors can be the last wall to start actions like reducing the access of cars to city centers, to mitigate the impact on the public health.
Among the different types of gas sensors, chemoresistive sensors have been attracting great interest due to their high sensitivity, low cost, reproducibility and simplicity [3]. Other techniques, like gas chromatography, electrochemical, and optical sensors, present some drawbacks, because they are more expensive, require trained personnel and still present some problems to miniaturize and export the measurement systems out of the laboratory [4,5]. However, chemoresistive sensors can be easily adapted to in-field measurements with low power consumption, especially when these sensors work at room temperature [6], reducing the complexity of the sensor driving circuitry. Nevertheless, Table 1. Comparison of the sensitivity for different chemoresistive sensors that employ perovskites. NT = not tested. NO 2 and NH 3 sensitivities expressed as 10 −3 % ppb −1 and 10 −3 % ppm −1 , respectively. Relative humidity (R.H.) effect expressed as the variation of the sensitivity. Degradation information about the stability of the perovskite during a few days or weeks. EDPIC = ethylenediamine lead iodide chloride.  [36] Threshold limit values (TLV) to 1-h exposure to NO 2 are established at 200 ppb and 100 ppb, by the European Union [37] and the Environmental Protection Agency of the US [38], respectively. Furthermore, the average yearly limit mean for NO 2 exposure is set at 40 ppb in the EU and at 53 ppb in the US. Maximum exposure values for NH 3 are defined as the ST (short-term) and TWA (Time-Weighted Average) by the European Agency for Safety and Health at Work (EU-OSHA) [39] and by the National Institute for Occupational Safety and Health (NIOSH) in the US [40]. These data are summarized in Table 2. TWA means the average exposure of 8 h/day and ST is referred to short exposures of 15 min. Both institutions define similar ST and TWA values for NH 3 limits. Thus, the development of sensor networks able to detect these trace concentration levels in a wide range of locations would be highly advisable. In consequence, sensors should offer reversible interaction with gas molecules for allowing the continuous monitoring of these species at an affordable cost.

Perovskite Synthesis and Graphene Preparation
To synthesize lead halide perovskite (MAPbBr 3 ) nanocrystals, the method proposed by Schmidt, et al. [41] was followed, in which a 2 mL solution of 1-octadecene (ODE) was prepared and 85 mg of oleic acid (OLA) were added. Then, the solution was stirred and heated up to 80 • C in a hotplate and 33.5 mg of octylammonium bromide (OABr) were added. Continuously, both, a solution of methyl-ammonium bromide (MABr) in dimethylformamide (DMF) and another with lead (II) bromide (PbBr 2 ) were mixed in the same solvent. Subsequently, these two solutions were added to the main solution and was cooled to 60 • C. Finally, 5 mL of acetone was added to the final solution to create precipitates of perovskite nanocrystals, which were separated from the unreactive solution by using a centrifugation technique at 6000 rpm during 10 min. Afterwards, the nanocrystals obtained were suspended in toluene.
A graphene solution was prepared using 1 mg of graphene nanoplatelets from Strem Chemicals, Inc. (USA) dispersed in 1 mL of toluene. Then, graphene nanoplatelets were sonicated employing a Sonic Tip (FisherbrandTM Model 705) at 40% of 700W for 1 h and 30 min using a 1 s on −2 s off pulsed sonication.

Material Characterization
The exfoliated graphene and the lead halide perovskite were characterized by several techniques such as XPS, HRTEM, FESEM and Raman spectroscopy.
The chemical composition of the nanomaterials obtained was studied via XPS, using a SPECS spectrometer (Berlin, Germany) equipped with a Phoibos 150 MCD detector, using a non-monochromatic X-ray source (Al) operating at 200 W. The intensity ratios of the different components were calculated from the area peak after a correction by the transition function of the spectrometer and a non-linear Shirley-type background subtraction. Additionally, graphene crystallinity was evaluated employing a Raman spectrometer from Renishaw, plc. (Wotton-under-Edge, UK), coupled to a confocal Leica DM2500 microscope. The laser used had a wavelength of 514 nm.
The morphology was studied employing a JEOL JEM 2100F (Tokio, Japan) HRTEM, at an operating voltage of 200 kV. Nanomaterials were ultrasonically dispersed in toluene and a drop of dispersion was deposited onto a carbon-coated copper grid, drying it at room temperature. Graphene porosity and perovskites NCs distribution were analyzed by Carl Zeiss AG -ULTRA 55 (Oberkochen, Germany) Field Emission Scanning Electron Microscope.

Sensor Fabrication
The materials obtained were deposited onto 1.5 cm 2 quartz substrates previously cleaned in an ultrasonic bath, using an anionic detergent (Alconox) and miliQ water. Afterwards, a 1M solution of hydrochloric acid was employed in order to increase the surface hydrophilicity. Finally, the substrates were cleaned and dried several times with ethanol. Then, a film of graphene nanoplatelets was deposited onto quartz substrates by drop casting method, and subsequently, perovskite nanocrystals (5% wt solution) were deposited by spin coating at 1000 rpm for 120 s.
Once the hybrid material film was deposited onto quartz substrate, its backside was glued with a thermally conducting epoxy from Heraeus, Inc. (Hanau, Germany) to an alumina hotplate that comprised a screen-printed platinum heater. Then, the samples were placed on a 20 × 30 mm printed circuit board (PCB) to be connected to an airtight test chamber. Finally, two-wire contacts were made on the sensor surface employing a conductive silver paste and platinum wires (a picture of a typical sensor is shown in the supporting information, see Figure S1).

Gas sensing Measurements
The resistance changes under different gases and experimental conditions were monitored using an Agilent HP 34972A multimeter connected to the gas sensing chamber. Additionally, calibrated gas cylinders were employed to apply synthetic air (Air Premier Purity: 99,995%) and different dilutions of gases tested to achieve the target concentration. In order to reproduce real atmospheric conditions, NO 2 and NH 3 gases were balanced in synthetic air as well. The total flow was adjusted to 100 mL/min using a set of Bronkhorst High-Tech B.V. (Ruurlo, The Netherlands) mass-flows controllers. The sensors were stabilized under synthetic dry air for 60 min before the application of the desired gas concentration during 30 min of exposure. Besides, a controller evaporator mixer (CEM) from Bronkhorst High-Tech B.V. (Ruurlo, The Netherlands) was used to humidify the gas mixture and thus, analyze the effect of ambient moisture on the sensing properties (see Figure S2). The relative humidity level was monitored by using a SHT51 humidity sensor from Sensirion AG (Stäfa, Switzerland) placed in a 35 cm 3 airtight chamber. The ambient temperature inside the test chamber throughout the measurement process was 23 • C ± 1 • C.

Material Characterization
A high-resolution transmission electron microscopy (HR-TEM) analysis was conducted to characterize graphene and perovskite nanocrystals independently. Figure 1a shows graphene layers with diameters near hundreds of nanometers. The interplanar distance obtained was 2.4 Angstrom (Figure 1b), showing a suitable graphene crystallinity. The perovskite synthesis method produces small nanocrystals, these lead halide perovskite nanocrystals to appear as dark spots (Figure 1c) with an average size of 7.1 ± 2.2 nm (see Figure S3 in the Supplementary Materials). Figure 1d shows the high crystallinity of the perovskites synthesized (Figure 2b), with a calculated interplanar distance of 2.8 Angstrom.  Additionally, FESEM images were obtained for both, bare and graphene decorated with MAPbBr3 NCs. Figure 2a shows the graphene once it was deposited onto the quartz substrates to be employed as a gas sensor, presenting a porous surface, which is interesting for gas sensing. Besides, in Figure 2b, the graphene decorated with perovskites nanocrystals can be observed, corresponding to the bright spots. To obtain this image, a Back-Scattered Electron detector (BSE) was used, revealing a quite homogeneous distribution of the nanocrystals on the surface of graphene. Additionally, FESEM images were obtained for both, bare and graphene decorated with MAPbBr 3 NCs. Figure 2a shows the graphene once it was deposited onto the quartz substrates to be employed as a gas sensor, presenting a porous surface, which is interesting for gas sensing. Besides, in Figure 2b, the graphene decorated with perovskites nanocrystals can be observed, corresponding to the bright spots. To obtain this image, a Back-Scattered Electron detector (BSE) was used, revealing a quite homogeneous distribution of the nanocrystals on the surface of graphene.
According to the XPS elemental quantification, the graphene used presents a 7.8% content of oxygen and 92.2% of carbon. The results of the XPS fitting analysis are represented in Figure 3. The C1s spectrum, centered at 284. 35  systems in the "graphite-like" [42]. The peak reproduced at 290.5 eV is also characteristic of sp 2 carbon systems, but in this case due to the π-π* interactions and π plasmon, which is derived from the energy loss due to the excitation of π electrons [43]. Components associated with carbon in amorphous or sp 3 configuration appear at 285.0 eV [44], and defects associated with carbon vacancies appear at 282.9eV [45]. Meanwhile, associated to carbon-oxygen bonds, three peaks can be observed at 286.5 eV, 288.1 eV and 289.0 eV, which correspond to C-O, C=O and carboxylic groups, respectively [46]. Peak quantification is summarized in Tables S1 and S2 (see the supporting information). According to the XPS elemental quantification, the graphene used presents a 7.8% content of oxygen and 92.2% of carbon. The results of the XPS fitting analysis are represented in Figure 3. The C1s spectrum, centered at 284.35 eV, is deconvoluted in seven peaks ( Figure 3a). The component at 284.2 eV is characteristic of photoelectrons emitted from carbon atoms, associated with sp 2 carbon systems in the "graphite-like" [42]. The peak reproduced at 290.5 eV is also characteristic of sp 2 carbon systems, but in this case due to the π-π* interactions and π plasmon, which is derived from the energy loss due to the excitation of π electrons [43]. Components associated with carbon in amorphous or sp 3 configuration appear at 285.0 eV [44], and defects associated with carbon vacancies appear at 282.9eV [45]. Meanwhile, associated to carbon-oxygen bonds, three peaks can be observed at 286.5 eV, 288.1 eV and 289.0 eV, which correspond to C-O, C=O and carboxylic groups, respectively [46]. Peak quantification is summarized in Tables S1 and S2 (see the supporting information).
(a) (b) Figure 3. Deconvolution of the C 1s (a) and O 1s (b) core level peak for graphene.
The O 1s spectrum is reproduced employing three components centered at different binding energies, which are 531.3 eV, 532.6 eV and 533.9 eV (Figure 3b). The first one at 531.3 eV is attributed to oxygen physically absorbed, meanwhile the two peaks centered at 532.3 eV is associated to C=O, O-C=O and isolated OH, the peaks at 533.9 eV should correspond to C-O, C-O-OH and C-OH groups  According to the XPS elemental quantification, the graphene used presents a 7.8% content of oxygen and 92.2% of carbon. The results of the XPS fitting analysis are represented in Figure 3. The C1s spectrum, centered at 284.35 eV, is deconvoluted in seven peaks (Figure 3a). The component at 284.2 eV is characteristic of photoelectrons emitted from carbon atoms, associated with sp 2 carbon systems in the "graphite-like" [42]. The peak reproduced at 290.5 eV is also characteristic of sp 2 carbon systems, but in this case due to the π-π* interactions and π plasmon, which is derived from the energy loss due to the excitation of π electrons [43]. Components associated with carbon in amorphous or sp 3 configuration appear at 285.0 eV [44], and defects associated with carbon vacancies appear at 282.9eV [45]. Meanwhile, associated to carbon-oxygen bonds, three peaks can be observed at 286.5 eV, 288.1 eV and 289.0 eV, which correspond to C-O, C=O and carboxylic groups, respectively [46]. Peak quantification is summarized in Tables S1 and S2 (see the supporting information). The O 1s spectrum is reproduced employing three components centered at different binding energies, which are 531.3 eV, 532.6 eV and 533.9 eV (Figure 3b). The first one at 531.3 eV is attributed to oxygen physically absorbed, meanwhile the two peaks centered at 532.3 eV is associated to C=O, O-C=O and isolated OH, the peaks at 533.9 eV should correspond to C-O, C-O-OH and C-OH groups [47]. These oxygenated species will help to anchor and stabilize the perovskites in the graphene layers, similarly to the typical decoration of carbon nanomaterials with nanoparticles of different The O 1s spectrum is reproduced employing three components centered at different binding energies, which are 531.3 eV, 532.6 eV and 533.9 eV (Figure 3b). The first one at 531.3 eV is attributed to oxygen physically absorbed, meanwhile the two peaks centered at 532.3 eV is associated to C=O, O-C=O and isolated OH, the peaks at 533.9 eV should correspond to C-O, C-O-OH and C-OH groups [47]. These oxygenated species will help to anchor and stabilize the perovskites in the graphene layers, similarly to the typical decoration of carbon nanomaterials with nanoparticles of different nature. Additional Raman Spectroscopy analysis was conducted to confirm the crystallinity reported before and after loading graphene with the perovskite (see Figure S4 Supplementary Materials).

Gas Sensing Results
Two different gases, nitrogen dioxide and ammonia, with high interest to be monitored due to the danger to human health derived from their exposure at certain levels were measured. Several dilutions were performed in order to apply different concentrations, especially under the threshold limits. Due to the different baseline levels of the two films studied, normalized responses are show for better comparison in Figure 4. Also, it can be expected that a slight baseline drift appears, given the difficulty to completely clean the surface from adsorbed gas molecules due to the room temperature working conditions. However, a temperature treatment was performed on a weekly basis, to counteract drift, in which the sensors are heated up to 80 • C during 1 h to desorb the gas molecules at the surface and regain the original baseline.
Two different gases, nitrogen dioxide and ammonia, with high interest to be monitored due to the danger to human health derived from their exposure at certain levels were measured. Several dilutions were performed in order to apply different concentrations, especially under the threshold limits. Due to the different baseline levels of the two films studied, normalized responses are show for better comparison in Figure 4. Also, it can be expected that a slight baseline drift appears, given the difficulty to completely clean the surface from adsorbed gas molecules due to the room temperature working conditions. However, a temperature treatment was performed on a weekly basis, to counteract drift, in which the sensors are heated up to 80 ˚C during 1 h to desorb the gas molecules at the surface and regain the original baseline.
Nitrogen dioxide detection was performed by applying repeated exposure and recovery cycles to increasing concentrations of the analyte considered. Figure 4a shows the response to 250, 500, 750, and 1000 ppb, in which graphene loaded with perovskite nanocrystals obtained higher responses compared to bare graphene. Due to the high responses obtained in this range of concentrations, especially for perovskite-loaded graphene, the concentrations measured were decreased in order to analyze the sensitivity under a few ppb of NO2. Figure 4b shows the responses obtained at 25 and 100 ppb, under room temperature conditions.   Figure 4 shows that response dynamics are slow and the saturation of the response is not achieved during the 30-min gas exposure. This drawback is often experienced with carbon nanomaterial chemoresistors, especially for room-temperature operated sensors. Slow response dynamics can be ameliorated by reducing the dead volume of the test chamber or by increasing the gas flow. Sensor response was computed as R(%) = 100*(R−R0)/R0, Where R0 is the resistance value immediately before gas exposure (i.e., under dry air), and R is the resistance immediately before the start of a new cleaning step (in other words, the last resistance value acquired while under exposure to NO2 or NH3). Figure 5 shows the calibration curves obtained for the whole range of NO2 concentrations measured. Graphene doped with perovskite nanocrystals presents a higher response (up to 3-fold) than bare graphene, even under a hundred ppb of NO2 exposure. Additionally, bare graphene presents a slight saturation above 500 ppb, decreasing the slope obtained and, in consequence, reducing sensitivity. Nitrogen dioxide detection was performed by applying repeated exposure and recovery cycles to increasing concentrations of the analyte considered. Figure 4a shows the response to 250, 500, 750, and 1000 ppb, in which graphene loaded with perovskite nanocrystals obtained higher responses compared to bare graphene. Due to the high responses obtained in this range of concentrations, especially for perovskite-loaded graphene, the concentrations measured were decreased in order to analyze the sensitivity under a few ppb of NO 2 . Figure 4b shows the responses obtained at 25 and 100 ppb, under room temperature conditions. Figure 4 shows that response dynamics are slow and the saturation of the response is not achieved during the 30-min gas exposure. This drawback is often experienced with carbon nanomaterial chemoresistors, especially for room-temperature operated sensors. Slow response dynamics can be ameliorated by reducing the dead volume of the test chamber or by increasing the gas flow. Sensor response was computed as R(%) = 100*(R−R 0 )/R 0 , Where R 0 is the resistance value immediately before gas exposure (i.e., under dry air), and R is the resistance immediately before the start of a new cleaning step (in other words, the last resistance value acquired while under exposure to NO 2 or NH 3 ). Figure 5 shows the calibration curves obtained for the whole range of NO 2 concentrations measured. Graphene doped with perovskite nanocrystals presents a higher response (up to 3-fold) than bare graphene, even under a hundred ppb of NO 2 exposure. Additionally, bare graphene presents a slight saturation above 500 ppb, decreasing the slope obtained and, in consequence, reducing sensitivity. A stability study of graphene loaded with perovskite nanocrystals was conducted (Figure 5b), revealing a slight decrease in the response towards 500 ppb of NO2 after six months of sensor operation. This result confirms the high stability of MAPbBr3 under ambient moisture and reactive gases, probably due to the nanocrystal structure of the perovskite. Zhu et al. [28] report the high stability of perovskites in nanocrystal structure over other alternatives. Besides, in our work, the perovskites are probably protected by the hydrophobic properties of graphene. The process and experimental conditions followed for NO2 detection, were applied as well for NH3 detection at ppm range (Figure 6a). Similarly, to the detection of nitrogen dioxide, ammonia measurements reveal a higher sensitivity for graphene decorated with perovskite nanocrystals (Figure 6b). Besides, meanwhile bare graphene presents a saturated response, decorated graphene shows a high potential to enhance the response under longer exposures to the analytes. However, an analysis of the steady-state reveals that several hours were not enough to get the saturation of the perovskite decorated graphene ( Figure S5) at room temperature. A stability study of graphene loaded with perovskite nanocrystals was conducted (Figure 5b), revealing a slight decrease in the response towards 500 ppb of NO 2 after six months of sensor operation. This result confirms the high stability of MAPbBr 3 under ambient moisture and reactive gases, probably due to the nanocrystal structure of the perovskite. Zhu et al. [28] report the high stability of perovskites in nanocrystal structure over other alternatives. Besides, in our work, the perovskites are probably protected by the hydrophobic properties of graphene.
The process and experimental conditions followed for NO 2 detection, were applied as well for NH 3 detection at ppm range (Figure 6a). Similarly, to the detection of nitrogen dioxide, ammonia measurements reveal a higher sensitivity for graphene decorated with perovskite nanocrystals (Figure 6b). Besides, meanwhile bare graphene presents a saturated response, decorated graphene shows a high potential to enhance the response under longer exposures to the analytes. However, an analysis of the steady-state reveals that several hours were not enough to get the saturation of the perovskite decorated graphene ( Figure S5) at room temperature. A stability study of graphene loaded with perovskite nanocrystals was conducted (Figure 5b), revealing a slight decrease in the response towards 500 ppb of NO2 after six months of sensor operation. This result confirms the high stability of MAPbBr3 under ambient moisture and reactive gases, probably due to the nanocrystal structure of the perovskite. Zhu et al. [28] report the high stability of perovskites in nanocrystal structure over other alternatives. Besides, in our work, the perovskites are probably protected by the hydrophobic properties of graphene. The process and experimental conditions followed for NO2 detection, were applied as well for NH3 detection at ppm range (Figure 6a). Similarly, to the detection of nitrogen dioxide, ammonia measurements reveal a higher sensitivity for graphene decorated with perovskite nanocrystals (Figure 6b). Besides, meanwhile bare graphene presents a saturated response, decorated graphene shows a high potential to enhance the response under longer exposures to the analytes. However, an analysis of the steady-state reveals that several hours were not enough to get the saturation of the perovskite decorated graphene ( Figure S5) at room temperature. Additionally, the sensitivity under dry and humid conditions was evaluated to both, bare and perovskite loaded graphene. Figure 7 shows an increase in the response for graphene to 500 ppb of NO 2 (about 30%) when moisture is present. However, graphene decorated with perovskite nanocrystals shows a very stable response with only a slight decrease (about 5%) when under humid conditions. Additionally, the sensitivity under dry and humid conditions was evaluated to both, bare and perovskite loaded graphene. Figure 7b shows an increase in the response for graphene to 500 ppb of NO2 (about 30%) when moisture is present. However, graphene decorated with perovskite nanocrystals shows a very stable response with only a slight decrease (about 5%) when under humid conditions.

Discussion
Graphene with oxygen-containing functional groups such as carbonyl, hydroxyl, epoxy, and ether groups exhibited enhanced hole transport characteristics due to its suitable work function [48,49]. In addition, it is well known that molecules adsorbed on the surface of graphene can change the local carrier concentration [50]. In this regard, exposure to an electron-withdrawing gas, like NO2 leads to a lower sensor resistance. Conversely, an electron-donating gas, such as NH3, decreases the hole concentration resulting in a higher sensor resistance. In films decorated with perovskite nanocrystals, the p-type behavior of the sensor is defined as well by the graphene nanolayers (these are the dominant carrier transport material), meanwhile the MAPbBr3 NCs are ambipolar charge transporters [23]. In other words, perovskites can act as a p-or n-type semiconductor depending on the interaction between the sensor surface and the nature of the gas (electron-donating or electron-withdrawing).
Bare graphene samples offer a response to both, NO2 and NH3, due to the interaction of these gases with the oxygenated defects and functional groups located on the graphene nanosheets. However, all the measurements were done at room temperature, obtaining limited sensitivity derived from the small transfer of electronic charge between graphene and adsorbed gas molecules. Nevertheless, from the gas sensing tests, it was derived that the presence of MAPbBr3 NCs decorating the graphene nanolayers was advantageous for detecting NO2 and NH3 at room temperature. In addition, surface trap sites at the perovskite have been demonstrated to act as active sites for the gas sensing process. It is reported that a net positive charge is formed at the surface of the perovskite due to the loss of bromine and undercoordination of the Pb atom, favouring the perovskite to be sensitive to the environmental gases [23]. Regarding the baseline level in dry air, bare graphene shows a lower resistance baseline (3-fold) than the decorated one with MAPbBr3 NCs. Upon the formation of the hybrid perovskite the overall resistance of hybrid films is higher than that of bare graphene due to the intrinsic electrical properties of the perovskite nanocrystals.
This enhancement in the response when graphene is decorated with MAPbBr3 NCs is associated to the creation of electron-hole pairs by the perovskites, once these are exposed to the gases (Figure 8). On the one hand, nitrogen dioxide is getting adsorbed on the graphene due to the interaction between the gas and the oxygen defects and functional groups of graphene. As a consequence, the hole concentration is increasing, resulting in a decrease in the resistance. Additionally, the perovskite NCs have a crucial role in the improvement of the sensitivity towards

Discussion
Graphene with oxygen-containing functional groups such as carbonyl, hydroxyl, epoxy, and ether groups exhibited enhanced hole transport characteristics due to its suitable work function [48,49]. In addition, it is well known that molecules adsorbed on the surface of graphene can change the local carrier concentration [50]. In this regard, exposure to an electron-withdrawing gas, like NO 2 leads to a lower sensor resistance. Conversely, an electron-donating gas, such as NH 3 , decreases the hole concentration resulting in a higher sensor resistance. In films decorated with perovskite nanocrystals, the p-type behavior of the sensor is defined as well by the graphene nanolayers (these are the dominant carrier transport material), meanwhile the MAPbBr 3 NCs are ambipolar charge transporters [23]. In other words, perovskites can act as a p-or n-type semiconductor depending on the interaction between the sensor surface and the nature of the gas (electron-donating or electron-withdrawing).
Bare graphene samples offer a response to both, NO 2 and NH 3 , due to the interaction of these gases with the oxygenated defects and functional groups located on the graphene nanosheets. However, all the measurements were done at room temperature, obtaining limited sensitivity derived from the small transfer of electronic charge between graphene and adsorbed gas molecules. Nevertheless, from the gas sensing tests, it was derived that the presence of MAPbBr 3 NCs decorating the graphene nanolayers was advantageous for detecting NO 2 and NH 3 at room temperature. In addition, surface trap sites at the perovskite have been demonstrated to act as active sites for the gas sensing process. It is reported that a net positive charge is formed at the surface of the perovskite due to the loss of bromine and undercoordination of the Pb atom, favouring the perovskite to be sensitive to the environmental gases [23]. Regarding the baseline level in dry air, bare graphene shows a lower resistance baseline (3-fold) than the decorated one with MAPbBr 3 NCs. Upon the formation of the hybrid perovskite the overall resistance of hybrid films is higher than that of bare graphene due to the intrinsic electrical properties of the perovskite nanocrystals.
This enhancement in the response when graphene is decorated with MAPbBr 3 NCs is associated to the creation of electron-hole pairs by the perovskites, once these are exposed to the gases (Figure 8). On the one hand, nitrogen dioxide is getting adsorbed on the graphene due to the interaction between the gas and the oxygen defects and functional groups of graphene. As a consequence, the hole concentration is increasing, resulting in a decrease in the resistance. Additionally, the perovskite NCs have a crucial role in the improvement of the sensitivity towards NO 2 , because the electron-hole pairs are separated, creating an interface between the graphene and the MAPbBr 3 . As a result, the excess of positive charges (holes) generated at perovskite NCs are transferred from the perovskite to the graphene sheet [51]. Therefore, the graphene film will suffer a large increase in the number of holes (and associated decrease in resistivity) when exposed to nitrogen dioxide, thanks to the presence of perovskite NCs. Inversely, when perovskite NCs initiate the creation of electron-hole pairs by the interaction with an electron-donating gas, in our case NH 3 , an excess of negative charges (electrons) are generated. As a consequence, perovskite NCs transfer electrons to graphene (and this is associated to an increase in resistivity), enhancing the response of the film to NH 3 due to the higher concentration of electrons.
the MAPbBr3. As a result, the excess of positive charges (holes) generated at perovskite NCs are transferred from the perovskite to the graphene sheet [51]. Therefore, the graphene film will suffer a large increase in the number of holes (and associated decrease in resistivity) when exposed to nitrogen dioxide, thanks to the presence of perovskite NCs. Inversely, when perovskite NCs initiate the creation of electron-hole pairs by the interaction with an electron-donating gas, in our case NH3, an excess of negative charges (electrons) are generated. As a consequence, perovskite NCs transfer electrons to graphene (and this is associated to an increase in resistivity), enhancing the response of the film to NH3 due to the higher concentration of electrons. Figure 8. Schematic illustration of perovskite decorated graphene sensor showing the mechanism proposed after interaction with different gases. Two adsorption processes are proposed, one at the graphene surface and another at the perovskite NCs. (A) During the exposure to an electron-donating gas, an excess of positive charges is neutralized at the defective perovskite surface and the local hole concentration of the p-type graphene is decreased, which results in an increase in film resistance. (B) During the exposure to an electron-withdrawing gas, positive charges (holes) in the NCs are formed, which are transferred to the graphene layers from the NCs, decreasing the overall resistance of the hybrid film.
One of the main drawbacks of the sensors based on perovskites films is the fast degradation derived from the exposure to ambient moisture. For that reason, a lead halide perovskite with bromide was chosen, due to the well-known stability over other perovskites based in halides like iodine [33]. Additionally, this paper reports perovskites in nanocrystal form, in front of the most usual approximation, based on films. With NCs, more stable perovskites can be obtained [28], enhancing the lifetime and durability of the sensor. Besides, the high hydrophobicity of the graphene can protect the MAPbBr3 NCs and slow down their degradation, due to the limited interaction with ambient moisture [30].
Considering the effect of ambient moisture in the sensing mechanism, higher sensitivity under humid conditions can be observed for bare graphene. This is in agreement with other works that report the use of graphene as humidity sensors [52]. Normally, water molecules act as electron-withdrawing, interacting with the oxygen functional groups grafted in the graphene. Besides, the detection of NO2 in humid conditions can be favored by a water-mediated adsorption, as reported in chemoresistive sensors [53]. One of the main drawbacks of the sensors based on perovskites films is the fast degradation derived from the exposure to ambient moisture. For that reason, a lead halide perovskite with bromide was chosen, due to the well-known stability over other perovskites based in halides like iodine [33]. Additionally, this paper reports perovskites in nanocrystal form, in front of the most usual approximation, based on films. With NCs, more stable perovskites can be obtained [28], enhancing the lifetime and durability of the sensor. Besides, the high hydrophobicity of the graphene can protect the MAPbBr 3 NCs and slow down their degradation, due to the limited interaction with ambient moisture [30].
Considering the effect of ambient moisture in the sensing mechanism, higher sensitivity under humid conditions can be observed for bare graphene. This is in agreement with other works that report the use of graphene as humidity sensors [52]. Normally, water molecules act as electron-withdrawing, interacting with the oxygen functional groups grafted in the graphene. Besides, the detection of NO 2 in humid conditions can be favored by a water-mediated adsorption, as reported in chemoresistive sensors [53].
Nevertheless, the detection of NO 2 in the presence of ambient moisture for MAPbBr 3 decorated graphene shows interesting characteristics. Figure 7 shows that a very similar response to 500 ppb of NO 2 is achieved for hybrid sensors under dry and humid conditions. The response is only slightly lower in the presence of ambient moisture. Probably water molecules partly passivate the charge transport and the creation of electron-hole pairs of perovskite nanocrystals [54], slightly decreasing the improvement in the response registered by the presence of MAPbBr 3 under dry conditions. The hydrophilic properties of perovskite decorated graphene were further studied by contact angle measurements (see Figure S6 in the Supplementary Materials).
The gas-sensitive layer developed shows its potential for detecting concentrations of nitrogen dioxide and ammonia at trace levels, even below their threshold limit values. These results are a promising beginning for the use of perovskite nanocrystal decorated graphene in gas sensing applications. However, selectivity should be further studied before their use under real conditions. One reasonable approach could be the development of multi-sensor arrays employing decorated graphene with different types of perovskite nanocrystals.

Conclusions
This work reports the first use of graphene loaded with perovskite nanocrystals as a chemoresistive sensor. This nanocomposite material has been shown to be a good option to obtain gas sensors with high stability over time, avoiding the main problem associated with perovskites, which is their degradation in contact with ambient moisture. Decorated graphene nanolayers show enhanced nitrogen dioxide and ammonia sensitivity in comparison to their bare graphene counterparts and much reduced cross-sensitivity to the ambient moisture. Furthermore, a reversible response to NO 2 and NH 3 was achieved at room temperature, enabling the integration of these hybrid films in low-power gas sensing devices. In summary, perovskites can constitute an alternative to metals, metal oxides, polymers or other molecules commonly used in the modification of the surface of carbon nanomaterials such as graphene in view of tuning sensitivity and selectivity.